Published by Winding AI, Inc., the company that makes Griffin. No vendor paid to be listed.Fees read November 19, 2026 · About this site
bestllmgateway.comSee the gateways

Best LLM gateway / Fees data 2026

Read on November 19, 2026

LLM gateway pricing in 2026: what 13 gateways add to your bill

The fee model of 13 LLM gateways, read on their own pages on November 19, 2026: Cloudflare AI Gateway pricing, LiteLLM pricing, Vercel AI Gateway pricing, Portkey pricing and nine more. Gateways charge in four ways (a cut of credits you buy, a fee per request or log, a seat price, or nothing because you run them), so every one is also priced on the same $10,000 month.

  • 5–8%cut on credits you buy from a hosted gateway; Vercel takes 0%
  • $0–$550fee on a $10,000 month of model spend
  • 3 of 13changed owner or name in 2026
  • 11 µs–40 msoverhead the vendors claim, each on its own test

Updated November 20, 2026. Download: llm-gateway-fees-2026.csv

The 13 gateways and how they charge

In this site's order: jobs done without an enterprise contract (one API, fallbacks, caching, budgets, rate limits, self-hosting), then the lower fee on the sample month (with your own keys, then when buying credits), then alphabetical.

GatewayJobsFee modelYour own keysFree tierPaid plansLicenceLogs kept
Bifrost6 of 6Free; Enterprise customBuilt inWhole open-source buildEnterprise, 14-day trialApache-2.0Yours
LiteLLM6 of 6Free; Enterprise annualBuilt inWhole open-source buildEnterprise, quoteMIT + enterprise/ dirYours
Cloudflare AI Gateway5 of 6Core free; 5% on Unified Billing creditsNo feeEvery planWorkers Paid from $5/moHosted onlyWorkers Logs: 3 days free, 7 paid
LLM Gateway5 of 65% on credits (+1.5% intl. cards)FreeFree forever, pay per tokenEnterprise from about $3,000/yrAGPLv3 + ee/ dirMetadata; payloads 30 days (Enterprise)
TrueFoundry5 of 6Per user + per requestPass-through3 users × 10k requestsPro $25/user, +$20 per 100kEnterprise only7 days free, 30 Pro
Agent Router4 of 6FreeBuilt inWhole open-source buildHosted by other firmsApache-2.0Yours
OpenRouter4 of 65.5% on credits (8% Business)Free to $25k/mo, then 5%25+ free models, 50 req/dayPay as you goHosted onlyPrompts off by default
Portkey4 of 6Plan + per logStored integrations10k logs, 3 daysPro $49 + $9 per 100k (docs)Gateway MIT3 days free, 30 Pro
Helicone4 of 6Plan + per request; credits 0%Yes10k requests, 7 daysPro $79, Team $799Gateway GPL-3.0, no 2026 release7 days free, 1 month Pro
Kong AI Gateway3 of 6Konnect from $25/mo + $200 per 1M req.Built inOpen-source Kong, 6 AI pluginsKonnect Plus; Enterprise customApache-2.0; AI features EnterpriseAnalytics 14 months
Vercel AI Gateway3 of 60% on tokens; add-ons priced apartFree with fallback onMonthly credit, amount not statedPay as you goHosted onlyDashboard 36 days
Requesty3 of 65% on model cost0% (llms.txt)200 req/day, free modelsPay as you goHosted onlyNot stated
Griffin0 of 6Watch $0; Route 5% of routed spendAlways own keysWatch planTest $20 packs; Route 5%Hosted; MIT SDKBodies 7 days

Swipe the table sideways to see every column.

The fee on a $10,000 month

The sample: $10,000 of model spend, 2 million requests (about half a cent each), one seat where seats are billed, list prices.

GatewayWith your own keysBuying its credits
Bifrost$0 + serversNo hosted plan
LiteLLM$0 + serversNo hosted plan
Cloudflare AI Gateway$0$500
LLM Gateway$0$500
TrueFoundryAbout $425No token resale
Agent Router$0 + serversNo hosted plan
OpenRouter$0$550
Portkey$220No token resale
HeliconeAbout $243 + storage0% (credits waitlisted)
Kong AI Gateway$0 open source; $225 KonnectNo token resale
Vercel AI Gateway$0$0
Requesty$0$500
Griffin$0 to watchDoesn't resell tokens

The arithmetic: TrueFoundry $25 + 20 blocks of 100,000 requests at $20. Portkey $49 + 19 × $9 for logs past 100,000. Helicone $79 + $164 on its request steps ($0.0007 down to $0.00004375 a request), plus storage. Kong Konnect Plus $25 + $200 for the second million requests. OpenRouter's own-key use is free up to $25,000 a month. LLM Gateway adds 1.5% on international cards. Griffin's Route fee applies only to spend it moves to a cheaper model: if $2,000 of the month is routed, about $100.

Cloudflare, LiteLLM, Vercel and Portkey pricing

Cloudflare AI Gateway pricing

Analytics, caching and rate limiting are free on every plan. Unified Billing adds 5% to credits you buy, with provider prices passed through. Logs: accounts whose first gateway came on or after 24 September 2026 use Workers Logs (200,000 a day kept 3 days on Free; 20 million a month kept 7 days on Paid, then $0.60 a million), moving to new pricing on 1 December 2026.

LiteLLM pricing

The open-source gateway is $0, “free forever”, self-hosted. Enterprise is annual, “sized to your usage” and “never per token”, with SSO free up to 5 users and a 30-day trial key.

Vercel AI Gateway pricing

No markup or platform fee on tokens. Own keys are free while fallback to Vercel's keys is on, $0.20 per 1,000 requests when off. Add-ons: team-wide allowlist or zero data retention $0.10 per 1,000 requests, regional hosting 5% of inference, trace drains $0.05 per 1,000 traces.

Portkey pricing

portkey.ai/pricing now redirects to a Palo Alto Networks sales form. The docs still list Developer free (10,000 logs, 3 days) and Pro at $49 a month for 100,000 logs, then $9 per 100,000 up to 3 million, with 30 days of logs.

Overhead each vendor claims

GatewayClaim and how it was measured
Bifrost11 µs per request at 5,000 RPS, mocked calls
LiteLLM8 ms p95 at 1,170 RPS, four instances, mock
Cloudflare AI GatewayNot stated
LLM GatewayNot stated
TrueFoundry+7 ms, one 1-CPU pod (docs)
Agent RouterNone for LLM calls
OpenRouterNo number
Portkey20–40 ms (docs)
HeliconeP95 under 5 ms (Rust repo)
Kong AI GatewayRelative only, vendor test
Vercel AI Gateway“Near-zero”
RequestyNot stated
GriffinWatch none; Route adds one edge hop

Owner and name changes in 2026

  • Portkey: Palo Alto Networks announced the deal on 30 April and closed it on 29 May 2026; it is sold as Prisma AIRS AI Gateway.
  • Helicone: Mintlify announced its purchase on 3 March 2026 and kept it in maintenance mode; since 12 November the site says it is “now part of Experiential Labs. Details to come soon.”
  • Envoy AI Gateway: renamed Agent Router and moved to the Agentic AI Foundation on 10 September 2026, same Apache-2.0 code.

Method

On November 19, 2026 we read each vendor's pricing page (and its llms.txt or Markdown copy where one exists), plan tables, docs on caching, fallbacks, budgets, rate limits and logs, and the licence file in each repo; stars and releases came from the GitHub API. A job counts only when it works without an enterprise contract. Griffin's row comes from griffin.app, its docs and terms; Griffin is made by Winding AI, Inc., which publishes this site.

What this data doesn't show

Model prices themselves, annual or volume discounts, enterprise quotes, the cost of running a self-hosted gateway, and whether a vendor's latency claim holds on your traffic. The 2-million-request month is an assumption; request-priced gateways cost more with smaller calls.

Cite this

Best LLM Gateway, “LLM gateway pricing in 2026: what 13 gateways add to your bill,” bestllmgateway.com, fees read November 19, 2026. https://bestllmgateway.com/llm-gateway-fees-2026/

Sources

Published by Winding AI, Inc., the company that makes Griffin. Griffin is one row of thirteen, under the same order rule. Corrections: support@winding.ai.